Abstract: Radiation shielding calculations for advanced nuclear systems are increasingly challenged by deep penetration, strong angular anisotropy, and complex three-dimensional geometries. Conventional discrete ordinate method can efficiently provide global flux solutions but may suffer from nonphysical ray effects, while Monte Carlo methods are flexible in geometry and physics modeling but often become statistically inefficient in low-flux regions. To address these difficulties, Zhejiang University has carried out systematic studies on the random ray method (TRRM) and developed the three-dimensional particle transport code THETA. TRRM retains the deterministic solution of the transport equation along characteristic rays, while replacing the fixed spatial-angular quadrature used in conventional characteristic methods with stochastic sampling. In each transport batch, new characteristic rays are generated with randomly sampled starting positions and flight directions. This formulation enables continuous statistical coverage of the angular domain and avoids long-term dependence on a fixed set of discrete directions. Meanwhile, because random rays are generated throughout the computational domain rather than only from the physical source region, low-flux regions can be sampled directly, which is particularly advantageous for deep-penetration shielding problems. Recent studies have investigated the performance of TRRM for both strongly anisotropic transport and deep-penetration problems. Numerical benchmarks show that TRRM can effectively suppress the ray effects observed in conventional characteristic calculations and produce smooth, physically reasonable flux distributions. In thick-shield and maze-type shielding problems, TRRM also maintains relatively stable statistical accuracy in regions where standard multigroup Monte Carlo calculations suffer from severe under-sampling. To further improve computational efficiency, several algorithmic developments have been introduced. Low-discrepancy sequences have been investigated to improve phase-space sampling, while a linear source approximation has been implemented to reduce the dependence on fine spatial meshes and thereby decrease memory and computational requirements. Based on these developments, THETA has been progressively verified and validated using theoretical benchmarks, shielding experiments, and reactor dosimetry problems. Applications to the VENUS-3 and PCA-Replica benchmarks demonstrate good agreement with reference and experimental results. More recently, THETA has been applied to the H.B. Robinson Unit 2 commercial pressurized-water-reactor benchmark, confirming its capability to model large-scale three-dimensional ex-core neutron transport over a very wide flux range. THETA is also being extended beyond the current structured-mesh TRRM framework. Ongoing developments include CAD-based unstructured-mesh transport, GPU parallelization using History-based and Event-based strategies, first-collision-source treatment for improved numerical stability, and forward/adjoint TRRM calculations for generating FW-CADIS weight windows. These developments indicate that THETA is evolving from a dedicated random-ray solver toward an integrated high-fidelity particle transport platform. Overall, the recent results demonstrate that TRRM provides a promising alternative for challenging radiation shielding problems by combining stochastic phase-space sampling with deterministic characteristic transport. With continued development in complex-geometry treatment, adaptive sampling, source-iteration acceleration, and heterogeneous computing, THETA is expected to support high-fidelity shielding analyses for advanced fission reactors, fusion systems, and other complex nuclear-engineering applications.